Portrait of Tong Zheng

Tong Zheng 郑童

zhengtong12356@gmail.com

I'm on the 2026–27 job market — please reach out for full-time research scientist roles starting Summer/Fall 2027.

About Me

I am a third-year Ph.D. student at the University of Maryland, College Park, advised by Prof. Heng Huang. Before my Ph.D., I was a research assistant at the Natural Language Processing Laboratory of Northeastern University (China), supervised by Prof. Tong Xiao, and received my B.E. in Computer Science from Northeastern University in 2021.

Research

I build scalable AI systems that close the discovery loop—systems that reason, explore, and recursively improve themselves.

Recursive Self-Improvement & Long-Horizon Discovery: Dream-RSI (arXiv'26) turns an agent's discovery history into a replay simulator it can dream in: a single costly online run then buys thousands of zero-execution-cost evaluations of alternative exploration strategies, making past experience the engine that recursively improves how the agent explores.

Parallel Reasoning & Test-Time Computation: Parallel-R1 (ICLR'26) is the first RL framework for parallel thinking in LLMs, moving beyond sequential chain-of-thought. Parallel-Probe (ICML'26) then makes parallel thinking efficient through 2D probing. AutoTTS (NeurIPS'26) unlocks a new direction for test-time scaling—shifting from hand-crafted inference heuristics to strategies the model discovers automatically via agentic search. MoT (ICLR'26) reasons over mixtures of complementary thought representations for logical reasoning.

Foundation Model Architectures: UMST (ICML'22) builds multiscale Transformers over sub-word, word, and phrase units with word-boundary and phrase-level structure. EIT (ACL'24) enhances multi-head self-attention by encouraging consensus across heads via inner- and cross-subspace interactions. PartialFormer (ACL Findings'24) replaces monolithic FFNs with multiple partial FFNs for parameter-efficient Transformers.

News

Older news
  • Check our new paper for VLM exploration: VOGUE — visual uncertainty guided exploration.
  • Check our new papers for LLM reasoning: Parallel-R1 and CDE.
  • One paper accepted for publication at EMNLP 2025.
  • Two papers accepted for publication at ACL 2025 Findings.
  • I will join Tencent AI Lab (Seattle) as a research intern this summer.
  • One paper accepted for publication at ICLR 2025.
  • One paper accepted for publication at NeurIPS 2024.
  • One paper accepted for publication at EMNLP 2024 Main.
  • Started my Ph.D. study at University of Maryland, College Park.
  • Two papers accepted for publication at ACL 2024 (1 Main, 1 Findings).
  • One paper accepted at Findings of EMNLP 2023.
  • Learning Multiscale Transformer Models for Sequence Generation accepted at ICML 2022 (First ICML in NEUNLP).
  • Joined NEUNLP lab as a research assistant.
  • Graduated from Northeastern University with an average GPA of 4.0.

Experience

Selected Publications (* Equal Contribution)

NEW

Dream-RSI: Recursive Self-Improvement through Evolving Worlds paper · website · code

arXiv, September 2026

Tong Zheng, Xidong Wu, Zheng Zhang, Zhankui He, Chaoyi Zhang, Benjamin Coleman, Ruoqiao Wei, Di Bai, Haolin Liu, Rui Liu, Xue Wang, Yue Zhuan, Wang-Cheng Kang, Renkai Xiang, Heng Huang, Xinwu Cheng, Yunsong Guo

NEW

LLMs Improving LLMs: Agentic Discovery for Test-Time Scaling paper · code

NeurIPS 2026

Tong Zheng, Haolin Liu, Chengsong Huang, Huiwen Bao, Sheng Zhang, Rui Liu, Runpeng Dai, Ruibo Chen, Chenxi Liu, Tianyi Xiong, Xidong Wu, Hongming Zhang, Heng Huang

NEW

Parallel-Probe: Towards Efficient Parallel Thinking via 2D Probing paper

ICML 2026

Tong Zheng*, Chengsong Huang*, Runpeng Dai*, Yun He, Rui Liu, Xin Ni, Huiwen Bao, Kaishen Wang, Hongtu Zhu, Jiaxin Huang, Furong Huang, Heng Huang

Parallel-R1: Towards Parallel Thinking via Reinforcement Learning paper · project · reported by VentureBeat · code

ICLR 2026 Main Conference; NeurIPS ER Workshop Spotlight

Tong Zheng, Hongming Zhang, Wenhao Yu, Xiaoyang Wang, Xinyu Yang, Runpeng Dai, Rui Liu, Huiwen Bao, Chengsong Huang, Heng Huang, Dong Yu

Learning to Reason via Mixture-of-Thought for Logical Reasoning paper

ICLR 2026 Main Conference; NeurIPS ER Workshop

Tong Zheng*, Lichang Chen*, Simeng Han, R. Thomas McCoy, Heng Huang

Towards Optimal Multi-draft Speculative Decoding paper

ICLR 2025

Zhengmian Hu*, Tong Zheng*, Vignesh Viswanathan, Ziyi Chen, Ryan A. Rossi, Yihan Wu, Dinesh Manocha, Heng Huang

A Bayesian Approach to Harnessing the Power of LLMs in Authorship Attribution paper

EMNLP 2024 Main

Zhengmian Hu*, Tong Zheng*, Heng Huang

PartialFormer: Modeling Part Instead of Whole Paper

ACL 2024 Findings

Tong Zheng*, Bei Li*, Huiwen Bao*, Weiqiao Shan, Tong Xiao, Jingbo Zhu

EIT: Enhanced Interactive Transformer Paper

ACL 2024 Main Conference

Tong Zheng*, Bei Li*, Huiwen Bao*, Tong Xiao, Jingbo Zhu

Learning Multiscale Transformer Models for Sequence Generation Paper

ICML 2022

Bei Li*, Tong Zheng*, Yi Jing*, Chengbo Jiao, Tong Xiao, Jingbo Zhu

Manuscript

BrainTGL: Temporal Graph Representation Learning for Brain Network by Exploiting Graph Temporal Information Manuscript

Finished at August 2021

Tong Zheng

Selected Honors